Agent Skills
On-demand skills for coding agents — UI audits, typography, docs, PR review, and releases — so you don't ship AI slop.
- Source repo
- mblode/agent-skills
- Stars
- ★ 215
- Last updated
- 1d ago
- License
- MIT
- Primary language
- TypeScript
- FA score
- 59/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code
- Cost
- Free software; you pay for model usage
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A frontend engineer runs ui-verification before merging a UI change, using nine headless-browser probes to replace subjective judgement with measured defects.
- Not a fit if
- Teams that use neither Claude Code nor Codex
- Users who want a pure GUI with no CLI or config
- Website
- blode.co/agent-skills ↗ set on GitHub
- Maintenance
- Commits in the last 30 dayslast commit 2026-11-19+2.0 ★/day over 7 days
- Source review
- 59/100 · Major gaps
What does this agent do, and when should you use it?
Agent Skills is a collection of on-demand skills for coding agents, maintained by Matthew Blode (blode.co) and published under the MIT license. Each skill lives as its own SKILL.md file and is installed into Codex and Claude Code with a single npx skills add command. The skills are grouped into six areas — Architecture, Design, Writing, Quality, Shipping, and Authoring — and include codebase-architecture, scaffold-nextjs, ui-design, ui-verification, typography-audit, tidy, pr-creator, pr-babysitter, autoship, agents-md, and more. Rather than running as a service, the repository ships instructions and rule sets that the agent loads when a matching task appears. Distribution goes through skills.sh, and every skill is linked to its own file in the repository.
The repository does not contain a standalone executable; it contains skill definition files such as skills/<name>/SKILL.md that tell an agent what steps and rules to follow for a given task. Skills are installed with npx skills add mblode/agent-skills -g --agent codex claude-code -y and then invoked by the agent during a session. Documented operations include: designing or hardening a code structure with codebase-architecture; scaffolding a Next.js turborepo with Blode UI, Ultracite and Vercel via scaffold-nextjs; booting an app in a headless browser and running nine probes to turn inferred UI defects into measured ones with ui-verification; checking punctuation, fonts, sizing, spacing, hierarchy and pairing against 78 rules with typography-audit; reviewing a diff or PR with file:line findings in confirmed and plausible tiers via tidy; creating, watching and releasing PRs with pr-creator, pr-babysitter and autoship; and wiring Claude Code, Codex and Cursor to a single set of instructions with agents-md.
- A frontend engineer runs ui-verification before merging a UI change, using nine headless-browser probes to replace subjective judgement with measured defects.
- A team maintaining an npm library runs dx-audit to score its library, CLI or SDK against 38 agent-friendliness rules.
- A writer or docs maintainer runs typography-audit to check punctuation, sizing and hierarchy against 78 rules.
- A developer uses agents-md to make Claude Code, Codex and Cursor read one shared AGENTS.md instead of drifting configs.
- A maintainer uses autoship to run the changesets-based npm release flow, covering the fix loop, CI watch and OIDC publish verification.
- A reviewer runs tidy on a PR to get file:line findings, optionally applying fixes and simplifying the diff.
Key setup facts
Pulled from the install and usage notes below. The project README has the current values.
- Install / run
npx skills add mblode/agent-skills -g --agent codex claude-code -y
How do you install or deploy this agent?
Prerequisites: Node.js with npx, a network connection, and an installed Codex or Claude Code agent. The README gives exactly one install command:
npx skills add mblode/agent-skills -g --agent codex claude-code -yThis installs globally through the skills.sh CLI and targets both the codex and claude-code agents. The README does not document alternative install paths (manual clone, offline install), required API keys, or permission scopes.
How do you use this agent?
After installation the skills live in the repository as SKILL.md files — the README links every skill to its own skills/<name>/SKILL.md — and the agent loads them on demand when a matching task appears. The runnable invocation shown in the README is the install command itself:
npx skills add mblode/agent-skills -g --agent codex claude-code -ySome skills are documented as explicitly named calls, for example test-audit (for a repo-wide or subsystem-wide pruning campaign) and app-verification ("Call by name"). The README does not show interactive invocation examples or per-skill parameters.
What are this agent's strengths and limitations?
- Skills are finely scoped with checkable rule counts — nine probes in ui-verification, 78 rules in typography-audit, 38 rules in dx-audit.
- One npx skills add command installs across two named agents, codex and claude-code, rather than locking you to a single runtime.
- Setup is a single command plus whatever the underlying agent already requires — no self-hosted service to stand up.
- Coverage spans the whole loop from scaffolding (scaffold-nextjs, scaffold-cli) to npm release (autoship), not just one stage.
- MIT licensed, with skills published as readable Markdown files that can be reviewed before use.
- The repository is skill definitions only; behaviour depends entirely on the host agent (Codex or Claude Code) and nothing runs standalone.
- Installation relies on the skills.sh CLI and network access, with no documented offline or manual path.
- The README does not state required API keys, permission scopes or runtime versions, so first-run cost is not fully predictable from the docs.
- The skill set is broad, but there are no shared invocation parameters or detailed per-skill usage examples.
- Some skills, such as test-audit and app-verification, must be explicitly named and are easy to overlook.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Agent Skills This agent | 59 · Major gaps | CLIFree + model costs | ★ 215 | 1d ago | TypeScript | Codex · Claude Code |
| Waza | 71 · Some gaps | CLIFree | ★ 7.6k | 3d ago | Python | Codex · Claude Code |
| Beautiful Rust | 71 · Some gaps | Agent plugin / skillFree + model costs | ★ 2.1k | 12d ago | Python | ChatGPT · Codex · Claude Code · Claude.ai |
| Loop Library: Reusable AI Agent Workflows | 75 · Good | Agent plugin / skillFree | ★ 3.2k | 4mo ago | JavaScript | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
The README offers global and single-skill installation and describes deployment through the skills CLI, supporting user control over scope; uninstall and permission details are limited, so least privilege and confirmation score 2 each. Workflows describe repository reads, validation, and public endpoint access, and some skills require network access; data-flow and sensitive-data guidance is partial, so each scores 1. Verification and website deployment workflows identify external effects, constraints, and failure conditions, but the collection has no uniform side-effect inventory, so external effects score 2. AGENTS.md gives a symlink removal method, and app-verification calls for isolation, providing limited recovery guidance; rollback scores 2. The README and license attribute the work to an author, but publisher identity is unverified, so source attribution scores 2.
AGENTS.md gives concrete installation-path and validator guidance, but README and guide install commands differ in options, and the sample skills do not consistently emphasize the same verification practices; self-consistency scores 1. Bash, curl, Node/npm, and Ruby/Python prerequisites are stated, but some skills depend on target-repository toolchains, external services, or reference files absent from this sample, so dependency availability scores 1. Workflows specify nonzero failures, actionable fixes, timeouts, and blocked outcomes, so failure messages score 2.
The README lists architecture, design, writing, quality, and shipping uses; sampled skills also define modes, inputs, and outputs, covering ordinary scenarios adequately for a 2. Several skills state scope exclusions and actions requiring authorization, so capability boundaries score 2. Descriptions include concrete invocation phrases and some name-only invocation guidance, so trigger precision scores 2. Compatibility fields and the guide explain host, runtime, and directory differences, though some host behavior depends on external tools; environment fit scores 2.
The README groups skills by category, and skill bodies provide contents, reference indexes, and workflows, so information architecture scores 2. README and AGENTS.md document installation, single-skill installation, development, and checks, scoring 2. Skills use kebab-case directories and matching names; sampled naming is stable, scoring 2. Samples provide commands, tables, checklists, and output contracts for ordinary use, scoring 2. They state exclusions, prerequisites, verification limits, and uncovered areas, scoring 2. The repository includes full MIT license text and copyright attribution, scoring 3. The supplied material shows no versioned release process or changelog, scoring 1. Maintenance guidance assigns upkeep tasks and describes the README, validator, and CI relationship, but publisher identity and long-term maintenance arrangements are limited, scoring 2.
Skills generally specify steps, deliverables, and completion conditions, making outputs actionable for a 2. The collection offers task-specific workflows, audit rules, and verification mechanisms that show marginal value beyond generic prompting, scoring 2. Installation is a one-time CLI operation and skills are invoked on demand; runtime and external-tool costs vary and are not quantified, so cost-benefit scores 2.
Workflows, commands, check conditions, and verification outputs are described concretely, supporting traceability for a 2. The README, AGENTS.md, sampled skills, CI, and tests corroborate some details, but the sample is incomplete and many claimed behaviors cannot be cross-checked against the supplied files, so cross-source corroboration scores 1. The material distinguishes scope, procedural facts, and judgment calls; product capability claims remain largely author assertions, so fact-inference separation scores 2.
- This assessment uses only the supplied static sample; unseen files are unknown, and installation, skill workflows, and tests were not run.
- README and AGENTS.md show different installation command options; users should check the command against their host and intended install scope.
- Some skills require network access, external services, or a target repository toolchain; the sample does not fully establish their data flows or availability.
FAQ
Does using these skills cost anything?
Do these skills work with agents other than Claude Code?
Are skills applied automatically once installed?
Will these skills modify my code?
What do the -g and -y flags in the install command do?
Is Agent Skills open source?
Is Agent Skills safe to use?
Is Agent Skills still maintained?
Who makes Agent Skills?
What is Agent Skills's official website?
Thanks — we use this to improve the page.